Andrew D. Straw: Closed-Loop, Visually-Based Flight Regulation in a Model Fruit Fly
This is a talk given at the Redwood Center for Theoretical Neuroscience, UC Berkeley on November 11, 2006 by Andrew Straw, Bioengineering, California Institute of Technology.
Run time 57:47Producer Redwood Center for Theoretical NeuroscienceAudio/Visual sound, colorContact Information Kilian Koepsell Redwood Center for Theoretical Neuroscience University of California Helen Wills Neuroscience Institute 132 Barker, MC #3190 Berkeley, CA 94720-3190
Abstract: Control theory provides a formal framework to understand feedback-based control, and is a tool neuroscientists may employ as they seek to understand animal behavior and physiology in a closed-loop context within the environment. Flight behavior of the fruit fly Drosophila, because of its robust and high-performance nature, its well-understood sensory and motor system constituent elements, and its amenability to multiple experimental approaches, represents an ideal system for investigating neural function within such a theoretical framework. Therefore, we are using control theory to investigate how the nervous system of a fly creates multiple levels of effective and robust flight behavior, from attitude stabilization, velocity regulation, obstacle avoidance, object search and localization, to landing. By implementing a physically realistic model of important components of the visual and motor systems of the fly, we are able to generate experimentally testable hypotheses about the control algorithms that may be implemented by the nervous and musclo-skeletal systems. Furthermore, questions about the function of individual components of these systems may be investigated at the theoretical level within the broader context of whole-animal function.
I will discuss in detail the visual system modeling I have performed, which is based upon a 3D environment model rendered by a computer graphics engine, spatially low-pass filtered and sampled to approximate the optics of a fly compound eye, and processed using simulated neurons consistent with our knowledge of fly visual physiology. In particular, motion detecting neurons with large receptive fields act as 'matched filters' for optical flow produced by forward and vertical translational velocity. Using estimates of these quantities produced by model neurons, a control algorithm for wingbeat kinematics is able to successfully follow a reference trajectory over a range of conditions. These results will be discussed in the context of behavioral experiments on freely flying flies which seek to test predictions from the present modeling results. Our experiences show that a control theory approach is useful for understanding closed-loop behavior of an animal within its environment because it enables us to create, and experimentally test, algorithms by which animals might control motor output to achieve a particular behavior.